CLI, Python and npm

Sign in with OAuth and use Quantura market search, forecasts, downloads and Scout.

Install and sign in

API access requires pay-as-you-go or Pro (including an active 14-day trial), enterprise or verified administrator access. The 14-day trial includes API keys, API calls, and MCP access.

npm install -g 'https://github.com/tamzid2001/stockssagemakerdata/releases/download/quantura-sdk-v1.0.1/quantura-sdk-1.0.1.tgz'
quantura login
quantura whoami
quantura search AAPL --source alpaca
quantura models

Python 3.10+ uses the release wheel:

python -m pip install 'https://github.com/tamzid2001/stockssagemakerdata/releases/download/quantura-sdk-v1.0.1/quantura_sdk-1.0.1-py3-none-any.whl'
python -m quantura_sdk login

Node 20+ and Python share the private credential file ~/.config/quantura/credentials.json. Choose one CLI per environment, or use python -m quantura_sdk. Browser sign-in uses your Quantura Google or email/password account through Clerk. PKCE S256, state and issuer checks protect the callback at http://127.0.0.1:8766/callback. quantura login --no-browser prints the sign-in URL. Tokens refresh automatically; quantura logout revokes the grant and removes local credentials.

For unattended integrations, use a personal key from Profile → Account settings → API keys and the server environment variable QUANTURA_API_KEY. Credentials are never command-line arguments.

Clients

import { Quantura } from 'quantura-sdk';
import { accessToken } from 'quantura-sdk/oauth'; // Node only

const q = new Quantura({ token: accessToken });
const results = await q.search('AAPL', { source: 'alpaca' });
const capabilities = await q.models();

from quantura_sdk import Quantura
from quantura_sdk.oauth import access_token

q = Quantura(access_token)
results = q.search("AAPL", source="alpaca")
capabilities = q.models()

Without an explicit token, clients read QUANTURA_API_KEY or QUANTURA_ACCESS_TOKEN. Responses retain the server’s data and meta envelopes. npm includes TypeScript declarations; Python includes its typing marker. Both clients have no runtime dependencies.

Create, poll and download

Inspect capabilities before selecting models or intervals. Save the request in request.json:

{
  "source": {"type": "ticker", "provider": "alpaca", "symbol": "AAPL", "frequency": "1Day", "limit": 500, "field": "close"},
  "prediction_length": 7,
  "horizon_mode": "trading_sessions",
  "quantiles": [0.01, 0.10, 0.25, 0.50, 0.75, 0.90, 0.99],
  "models": {"prophet": {"enabled": true, "weight": 1}},
  "failure_policy": "renormalize"
}

quantura forecast --file request.json --key aapl-daily-001
quantura get FORECAST_ID --wait
quantura download FORECAST_ID --output forecast.csv

created = q.create_forecast(request, idempotency_key="aapl-daily-001")
forecast_id = created["data"]["forecast_id"]
completed = q.wait_for_forecast(forecast_id)

The npm equivalents are createForecast, waitForForecast and downloadForecast. Preserve an explicit idempotency key when retrying after an uncertain transport failure. Clients never automatically retry job creation. Poll the returned ID instead of submitting another forecast. Cutoff fields such as history_lag_minutes are preserved; historical forecasts explain provider availability.

Paged Dukascopy downloads

request = {"source": "dukascopy", "symbol": "XAUUSD", "timeframe": "1Hour", "price_side": "bid", "start": "2025-10-08", "end": "2026-10-07"}
for page in q.history_pages(request):
    print(page)

The npm equivalent is for await (const page of q.historyPages(request)). The iterator follows next_cursor, preserves settings and rejects cursor loops. CLI history --file history.json --all-pages --output pages.json saves page envelopes. Rows are not silently filled, resampled or flattened.

Ask Scout

answer = q.ask_scout({"kind": "ensemble", "id": forecast_id}, "Where is the widest forecast range?")

Use q.askScout in npm. Scout grounds structured answers in authorized forecast, game, CSV preview or SageMaker context. Questions save a conversation/request in your account. See context and response schemas.

Errors and authorization

Clients expose safe error codes, HTTP status and request references. HTTPS is required outside loopback development; redirects are refused. OAuth credentials are bound to Quantura. Every request checks current identity, paid access, scopes and workspace permissions. These clients do not place trades.

Connect ChatGPT through MCP · npm release package · Python release

Unified provider history

history() uses the shared Download route. For Gemini and economic datasets, send a structured source object; use describe(source) to inspect economic fields and dimensions. The same source can create a forecast. MCP offers quantura_describe_dataset and quantura_download_history.

source = {"type": "gemini_spot", "symbol": "BTCUSD", "frequency": "1h", "limit": 500}
data = client.history({"source": source})